hermes-chatto-plugin/vendor/connectrpc/_server_sync.py

from __future__ import annotations

import base64
import functools
import traceback
from abc import ABC, abstractmethod
from dataclasses import replace
from http import HTTPStatus
from typing import TYPE_CHECKING, TypeVar
from urllib.parse import parse_qs

from . import _server_shared
from ._codec import Codec, get_default_codecs
from ._compression import negotiate_compression, resolve_compressions
from ._envelope import EnvelopeReader, EnvelopeWriter
from ._interceptor_sync import (
    BidiStreamInterceptorSync,
    ClientStreamInterceptorSync,
    InterceptorSync,
    MetadataInterceptorInvokerSync,
    MetadataInterceptorSync,
    ServerStreamInterceptorSync,
    UnaryInterceptorSync,
)
from ._protocol import ConnectWireError, HTTPException, ServerProtocol
from ._protocol_connect import (
    CONNECT_UNARY_CONTENT_TYPE_PREFIX,
    ConnectServerProtocol,
    codec_name_from_content_type,
)
from ._protocol_server import negotiate_server_protocol
from ._server_shared import (
    EndpointBidiStreamSync,
    EndpointClientStreamSync,
    EndpointServerStreamSync,
    EndpointUnarySync,
)
from .code import Code
from .errors import ConnectError
from .request import Headers, RequestContext

if TYPE_CHECKING:
    import sys
    from collections.abc import Callable, Iterable, Iterator, Mapping, Sequence
    from io import BytesIO

    if sys.version_info >= (3, 11):
        from wsgiref.types import ErrorStream, StartResponse, WSGIEnvironment
    else:
        from _typeshed.wsgi import ErrorStream, StartResponse, WSGIEnvironment

    from .compression import Compression
else:
    StartResponse = "wsgiref.types.StartResponse"
    WSGIEnvironment = "wsgiref.types.WSGIEnvironment"


_REQ = TypeVar("_REQ")
_RES = TypeVar("_RES")

_BODY_CHUNK_SIZE = 4096

# While _server_shared.EndpointSync is a closed type, we can't indicate that to Python so define
# a more precise type here.
EndpointSync = (
    EndpointBidiStreamSync[_REQ, _RES]
    | EndpointClientStreamSync[_REQ, _RES]
    | EndpointServerStreamSync[_REQ, _RES]
    | EndpointUnarySync[_REQ, _RES]
)


def _process_headers(environ: WSGIEnvironment) -> Headers:
    headers = Headers()
    if "CONTENT_TYPE" in environ:
        headers["content-type"] = environ["CONTENT_TYPE"]
    if "CONTENT_LENGTH" in environ:
        headers["content-length"] = environ["CONTENT_LENGTH"]

    for key, value in environ.items():
        if key.startswith("HTTP_"):
            header = key[5:].replace("_", "-")
            headers[header] = value
    return headers


def prepare_response_headers(
    base_headers: dict[str, list[str]], selected_encoding: str
) -> dict[str, list[str]]:
    """Prepare response headers with the selected compression encoding.

    Args:
        base_headers: Base response headers.
        selected_encoding: Selected compression encoding.

    Returns:
        The final response headers with content-encoding set.
    """
    headers = base_headers.copy()

    if "content-type" not in headers:
        headers["content-type"] = ["application/proto"]

    headers["content-encoding"] = [selected_encoding]
    headers["vary"] = ["Accept-Encoding"]
    return headers


def _read_body_with_content_length(
    environ: WSGIEnvironment, content_length: int
) -> bytes:
    input_stream: BytesIO = environ["wsgi.input"]

    # Many app servers buffer the entire request before executing the app
    # so do an optimistic read before looping.
    chunk = input_stream.read(content_length)
    if len(chunk) == content_length:
        return chunk

    bytes_read = len(chunk)
    chunks = [chunk]
    while bytes_read < content_length:
        to_read = content_length - bytes_read
        chunk = input_stream.read(to_read)
        if not chunk:
            break
        chunks.append(chunk)
        bytes_read += len(chunk)
    if bytes_read < content_length:
        raise ConnectError(
            Code.INVALID_ARGUMENT,
            f"request truncated, expected {content_length} bytes but only received {bytes_read} bytes",
        )
    return b"".join(chunks)


def _read_body(environ: WSGIEnvironment) -> Iterator[bytes]:
    input_stream: BytesIO = environ["wsgi.input"]
    while True:
        chunk = input_stream.read(_BODY_CHUNK_SIZE)
        if not chunk:
            return
        yield chunk


class ConnectWSGIApplication(ABC):
    """A WSGI application for the Connect protocol."""

    @property
    @abstractmethod
    def path(self) -> str: ...

    def __init__(
        self,
        *,
        endpoints: Mapping[str, EndpointSync],
        interceptors: Iterable[InterceptorSync] = (),
        read_max_bytes: int | None = None,
        compressions: Iterable[Compression] | None = None,
        codecs: Iterable[Codec] | None = None,
    ) -> None:
        """Initialize the WSGI application.

        Args:
            endpoints: A mapping of URL paths to endpoints. Typically provided directly
                by generated code from the Connect Python plugin.
            interceptors: A sequence of interceptors to apply to the endpoints.
            read_max_bytes: Maximum size of request messages.
            compressions: Supported compression algorithms. If unset, defaults to gzip.
                          If set to empty, disables compression.
            codecs: The codecs supported by the server. If unset, defaults to Protocol Buffers
                    binary and JSON codecs.
        """
        super().__init__()
        if interceptors:
            interceptors = [
                MetadataInterceptorInvokerSync(interceptor)
                if isinstance(interceptor, MetadataInterceptorSync)
                else interceptor
                for interceptor in interceptors
            ]
            endpoints = {
                path: _apply_interceptors(endpoint, interceptors)
                for path, endpoint in endpoints.items()
            }
        self._endpoints = endpoints
        self._read_max_bytes = read_max_bytes
        self._compressions = resolve_compressions(compressions)
        codecs = codecs if codecs is not None else get_default_codecs()
        self._codecs = {codec.name(): codec for codec in codecs}

    def __call__(
        self, environ: WSGIEnvironment, start_response: StartResponse
    ) -> Iterable[bytes]:
        ctx: RequestContext | None = None
        try:
            path = environ["PATH_INFO"]
            if not path:
                path = "/"
            endpoint = self._endpoints.get(path)
            if not endpoint and environ["SCRIPT_NAME"] == self.path:
                # The application was mounted at the service's path so we reconstruct
                # the full URL.
                endpoint = self._endpoints.get(self.path + path)

            if not endpoint:
                raise HTTPException(HTTPStatus.NOT_FOUND, [])

            http_method = environ["REQUEST_METHOD"]
            http_scheme = environ.get("wsgi.url_scheme", "http")
            headers = _process_headers(environ)
            if ra := environ.get("REMOTE_ADDR"):
                port = environ.get("REMOTE_PORT", "0")
                client_address = f"{ra}:{port}"
            else:
                client_address = None

            content_type = headers.get("content-type", "")
            protocol = negotiate_server_protocol(content_type)
            send_trailers: Callable[[list[tuple[str, str]]], None] | None = None

            if protocol.uses_trailers():
                send_trailers = environ.get("wsgi.ext.http.send_trailers")
                if not send_trailers:
                    msg = f"WSGI server does not support WSGI trailers extension but protocol for content-type '{content_type}' requires trailers"
                    raise RuntimeError(msg)
            ctx = protocol.create_request_context(
                endpoint.method, http_method, http_scheme, headers, client_address
            )

            if isinstance(endpoint, EndpointUnarySync) and isinstance(
                protocol, ConnectServerProtocol
            ):
                return self._handle_unary(
                    environ, start_response, http_method, endpoint, ctx, headers
                )
            return self._handle_stream(
                environ, start_response, send_trailers, protocol, headers, endpoint, ctx
            )

        except Exception as e:
            _drain_request_body(environ)
            _maybe_log_exception(environ, e)
            return self._handle_error(e, ctx, start_response)

    def _handle_unary(
        self,
        environ: WSGIEnvironment,
        start_response: StartResponse,
        http_method: str,
        endpoint: EndpointUnarySync[_REQ, _RES],
        ctx: RequestContext[_REQ, _RES],
        headers: Headers,
    ) -> Iterable[bytes]:
        # Handle request based on method
        if http_method == "GET":
            request, codec = self._handle_get_request(environ, endpoint)
        else:
            request, codec = self._handle_post_request(environ, endpoint, headers)

        # Process request
        response = endpoint.function(request, ctx)

        # Encode response
        res_bytes = codec.encode(response)
        base_headers = {
            "content-type": [f"{CONNECT_UNARY_CONTENT_TYPE_PREFIX}{codec.name()}"]
        }

        # Handle compression if accepted
        accept_encoding = headers.get("accept-encoding", "identity")
        compression = negotiate_compression(accept_encoding, self._compressions)
        res_bytes = compression.compress(res_bytes)
        response_headers = prepare_response_headers(base_headers, compression.name())

        # Convert headers to WSGI format
        wsgi_headers: list[tuple[str, str]] = []
        for key, values in response_headers.items():
            normalized_key = key.lower()
            wsgi_headers.extend((normalized_key, value) for value in values)
        _add_context_headers(wsgi_headers, ctx)

        start_response("200 OK", wsgi_headers)
        return [res_bytes]

    def _handle_post_request(
        self,
        environ: WSGIEnvironment,
        endpoint: _server_shared.EndpointSync[_REQ, _RES],
        request_headers: Headers,
    ) -> tuple[_REQ, Codec]:
        """Handle POST request with body."""

        codec_name = codec_name_from_content_type(
            request_headers.get("content-type", ""), stream=False
        )
        codec = self._codecs.get(codec_name)
        if not codec:
            raise HTTPException(
                HTTPStatus.UNSUPPORTED_MEDIA_TYPE,
                [("Accept-Post", "application/json, application/proto")],
            )

        try:
            content_length = environ.get("CONTENT_LENGTH")
            content_length = 0 if not content_length else int(content_length)
            if content_length > 0:
                req_body = _read_body_with_content_length(environ, content_length)
            else:
                req_body = b"".join(_read_body(environ))

            # Handle compression if specified
            compression_name = environ.get("HTTP_CONTENT_ENCODING", "identity").lower()
            if compression_name != "identity":
                compression = self._compressions.get(compression_name)
                if not compression:
                    raise ConnectError(
                        Code.UNIMPLEMENTED,
                        f"unknown compression: '{compression_name}': supported encodings are {', '.join(self._compressions.keys())}",
                    )
                try:
                    req_body = compression.decompress(req_body)
                except Exception as e:
                    raise ConnectError(
                        Code.INVALID_ARGUMENT,
                        f"Failed to decompress request body: {e!s}",
                    ) from e

            if (
                self._read_max_bytes is not None
                and len(req_body) > self._read_max_bytes
            ):
                raise ConnectError(
                    Code.RESOURCE_EXHAUSTED,
                    f"message is larger than configured max {self._read_max_bytes}",
                )

            try:
                return codec.decode(req_body, endpoint.method.input), codec
            except Exception as e:
                raise ConnectError(
                    Code.INVALID_ARGUMENT, f"Failed to decode request body: {e!s}"
                ) from e

        except Exception as e:
            if not isinstance(e, ConnectError):
                raise ConnectError(
                    Code.INTERNAL,
                    str(e),  # TODO
                ) from e
            raise

    def _handle_get_request(
        self, environ: WSGIEnvironment, endpoint: EndpointUnarySync[_REQ, _RES]
    ) -> tuple[_REQ, Codec]:
        """Handle GET request with query parameters."""
        try:
            query_string = environ.get("QUERY_STRING", "")
            params = parse_qs(query_string, keep_blank_values=True)

            if "message" not in params:
                raise ConnectError(
                    Code.INVALID_ARGUMENT,
                    "'message' parameter is required for GET requests",
                )

            message = params["message"][0]

            if "base64" in params and params["base64"][0] == "1":
                try:
                    message = base64.urlsafe_b64decode(message + "===")
                except Exception as e:
                    raise ConnectError(
                        Code.INVALID_ARGUMENT, f"Invalid base64 encoding: {e!s}"
                    ) from e
            else:
                message = message.encode("utf-8")

            # Handle compression if specified
            if "compression" in params:
                compression_name = params["compression"][0]
                compression = self._compressions.get(compression_name)
                if not compression:
                    raise ConnectError(
                        Code.UNIMPLEMENTED,
                        f"unknown compression: '{compression_name}': supported encodings are {', '.join(self._compressions.keys())}",
                    )
                message = compression.decompress(message)

            codec_name = params.get("encoding", ("",))[0]
            codec = self._codecs.get(codec_name)
            if not codec:
                raise ConnectError(
                    Code.UNIMPLEMENTED, f"invalid message encoding: '{codec_name}'"
                )
            # Handle GET request with proto decoder
            try:
                # TODO - Use content type from queryparam
                request = codec.decode(message, endpoint.method.input)
                return request, codec
            except Exception as e:
                raise ConnectError(
                    Code.INVALID_ARGUMENT, f"Failed to decode message: {e!s}"
                ) from e

        except Exception as e:
            if not isinstance(e, ConnectError):
                raise ConnectError(Code.INTERNAL, str(e)) from e
            raise

    def _handle_stream(
        self,
        environ: WSGIEnvironment,
        start_response: StartResponse,
        send_trailers: Callable[[list[tuple[str, str]]], None] | None,
        protocol: ServerProtocol,
        headers: Headers,
        endpoint: EndpointSync[_REQ, _RES],
        ctx: RequestContext[_REQ, _RES],
    ) -> Iterable[bytes]:
        req_compression, resp_compression = protocol.negotiate_stream_compression(
            headers, self._compressions
        )

        codec_name = protocol.codec_name_from_content_type(
            headers.get("content-type", ""), stream=True
        )
        codec = self._codecs.get(codec_name)
        if not codec:
            raise HTTPException(
                HTTPStatus.UNSUPPORTED_MEDIA_TYPE,
                [
                    (
                        "Accept-Post",
                        "application/connect+json, application/connect+proto",
                    )
                ],
            )
        writer = protocol.create_envelope_writer(codec, resp_compression)
        try:
            if not req_compression:
                raise ConnectError(
                    Code.UNIMPLEMENTED, "Unrecognized request compression"
                )
            request_stream = _request_stream(
                environ,
                endpoint.method.input,
                codec,
                req_compression,
                self._read_max_bytes,
            )
            match endpoint:
                case _server_shared.EndpointUnarySync():
                    request = _consume_single_request(request_stream)
                    response = endpoint.function(request, ctx)
                    response_stream = iter([response])
                case _server_shared.EndpointClientStreamSync():
                    response = endpoint.function(request_stream, ctx)
                    response_stream = iter([response])
                case _server_shared.EndpointServerStreamSync():
                    request = _consume_single_request(request_stream)
                    response_stream = endpoint.function(request, ctx)
                case _server_shared.EndpointBidiStreamSync():
                    response_stream = endpoint.function(request_stream, ctx)

            # Trigger service logic by consuming the first (possibly only) response message.
            first_response = next(response_stream, None)
            # Response headers set before the first message should be set to the context and
            # we can send them.
            _send_stream_response_headers(
                start_response, protocol, codec, resp_compression.name(), ctx
            )
            if first_response is None:
                # It's valid for a service method to return no messages, finish the response
                # without error.
                return [
                    _end_response(
                        writer.end(ctx.response_trailers, None), send_trailers
                    )
                ]

            # WSGI requires start_response to be called before returning the body iterator.
            # This means we cannot call yield in this function since the function would not
            # run at all until the iterator is consumed, meaning start_response wouldn't have
            # been called in time. So we return the response stream as a separate generator
            # function. This means some duplication of error handling.
            return _response_stream(
                first_response, environ, response_stream, writer, send_trailers, ctx
            )
        except Exception as e:
            # Exception before any response message was returned. An error after the first
            # response message will be handled by _response_stream, so here we have a
            # full error-only response.
            _drain_request_body(environ)
            _maybe_log_exception(environ, e)
            _send_stream_response_headers(
                start_response, protocol, codec, resp_compression.name(), ctx
            )
            return [
                _end_response(
                    writer.end(
                        ctx.response_trailers, ConnectWireError.from_exception(e)
                    ),
                    send_trailers,
                )
            ]

    def _handle_error(
        self, exc: Exception, ctx: RequestContext | None, start_response: StartResponse
    ) -> Iterable[bytes]:
        """Handle and log errors with detailed information."""
        headers: list[tuple[str, str]]
        body: list[bytes]
        status: str
        if isinstance(exc, HTTPException):
            headers = exc.headers
            body = []
            status = f"{exc.status.value} {exc.status.phrase}"
        else:
            wire_error = ConnectWireError.from_exception(exc)
            http_status = wire_error.to_http_status()
            headers = [("Content-Type", "application/json")]
            body = [wire_error.to_json_bytes()]
            status = f"{http_status.code} {http_status.reason}"

        if ctx:
            _add_context_headers(headers, ctx)

        start_response(status, headers)
        return body


def _end_response(
    message: bytes | Headers,
    send_trailers: Callable[[list[tuple[str, str]]], None] | None,
) -> bytes:
    if isinstance(message, bytes):
        return message
    assert send_trailers is not None  # noqa: S101
    send_trailers(list(message.allitems()))
    return b""


def _add_context_headers(headers: list[tuple[str, str]], ctx: RequestContext) -> None:
    headers.extend((key, value) for key, value in ctx.response_headers.allitems())
    headers.extend(
        (f"trailer-{key}", value) for key, value in ctx.response_trailers.allitems()
    )


def _send_stream_response_headers(
    start_response: StartResponse,
    protocol: ServerProtocol,
    codec: Codec,
    compression_name: str,
    ctx: RequestContext,
) -> None:
    response_headers = [
        ("content-type", protocol.content_type(codec)),
        (protocol.compression_header_name(), compression_name),
    ]
    response_headers.extend(
        (key, value) for key, value in ctx.response_headers.allitems()
    )
    start_response("200 OK", response_headers)


def _request_stream(
    environ: WSGIEnvironment,
    request_class: type[_REQ],
    codec: Codec,
    compression: Compression,
    read_max_bytes: int | None = None,
) -> Iterator[_REQ]:
    reader = EnvelopeReader(request_class, codec, compression, read_max_bytes)
    for chunk in _read_body(environ):
        yield from reader.feed(chunk)


def _response_stream(
    first_response: _RES,
    environ: WSGIEnvironment,
    response_stream: Iterator[_RES],
    writer: EnvelopeWriter,
    send_trailers: Callable[[list[tuple[str, str]]], None] | None,
    ctx: RequestContext,
) -> Iterable[bytes]:
    error: Exception | None = None
    try:
        body = writer.write(first_response)
        yield body
        for message in response_stream:
            body = writer.write(message)
            yield body
    except Exception as e:
        error = e
        _drain_request_body(environ)

    yield _end_response(
        writer.end(
            ctx.response_trailers,
            ConnectWireError.from_exception(error) if error else None,
        ),
        send_trailers,
    )


def _consume_single_request(stream: Iterator[_REQ]) -> _REQ:
    req = None
    for message in stream:
        if req is not None:
            raise ConnectError(
                Code.UNIMPLEMENTED, "unary request has multiple messages"
            )
        req = message
    if req is None:
        raise ConnectError(Code.UNIMPLEMENTED, "unary request has zero messages")
    return req


def _apply_interceptors(
    endpoint: EndpointSync[_REQ, _RES], interceptors: Sequence[InterceptorSync]
) -> EndpointSync:
    match endpoint:
        case EndpointUnarySync():
            func = endpoint.function
            for interceptor in reversed(interceptors):
                if not isinstance(interceptor, UnaryInterceptorSync):
                    continue
                func = functools.partial(interceptor.intercept_unary_sync, func)
            return replace(endpoint, function=func)
        case EndpointClientStreamSync():
            func = endpoint.function
            for interceptor in reversed(interceptors):
                if not isinstance(interceptor, ClientStreamInterceptorSync):
                    continue
                func = functools.partial(interceptor.intercept_client_stream_sync, func)
            return replace(endpoint, function=func)
        case EndpointServerStreamSync():
            func = endpoint.function
            for interceptor in reversed(interceptors):
                if not isinstance(interceptor, ServerStreamInterceptorSync):
                    continue
                func = functools.partial(interceptor.intercept_server_stream_sync, func)
            return replace(endpoint, function=func)
        case EndpointBidiStreamSync():
            func = endpoint.function
            for interceptor in reversed(interceptors):
                if not isinstance(interceptor, BidiStreamInterceptorSync):
                    continue
                func = functools.partial(interceptor.intercept_bidi_stream_sync, func)
            return replace(endpoint, function=func)


def _drain_request_body(environ: WSGIEnvironment) -> None:
    if environ.get("SERVER_PROTOCOL", "").startswith("HTTP/1"):
        # In HTTP/1, the request body should be drained before returning. Generally it's
        # best for the application server to handle this, but gunicorn is a famous
        # server that doesn't do so, so we go ahead and do it ourselves.
        for _ in _read_body(environ):
            pass


def _maybe_log_exception(environ: WSGIEnvironment, exc: Exception) -> None:
    if isinstance(exc, (ConnectError, HTTPException)):
        return
    errors: ErrorStream = environ["wsgi.errors"]
    errors.write(
        f"Exception in WSGI application\n{''.join(traceback.format_exception(type(exc), exc, exc.__traceback__))}"
    )